On Thu, Jan 8, 2009 at 10:41 AM, Ed Porter <ewpor...@msn.com> wrote: > ====Ed Porter====> > > This is certainly not true of a Novamente-type system, at least as I > conceive of it being built on the type of massively parallel, highly > interconnected hardware that will be available to AI within 3-7 years. Such > a system would be hierarchical in both the compositional and > generalizational dimensions, and the computation would be taking place by > importance weighted probabilisitic spreading activation, constraint > relaxation, and k-winner take all competition across multiple layers of > these hierarchies, so the decision making would not "funnel all reasoning > through a single narrowly focused process" any more that human though > processes do. > > > If a decision is to be made, it makes computational sense to have some > selection process that focuses attention on a selected one of multiple > possible candidate actions or > > though. If that is the type of "funneling" that you object to, you are > largely objecting to decision making itself.
I have been busy and I just started reading the remarks on this thread. I want to reply to Ed's comment since his remarks seemed to be focused in on what I said. (And I was able to understand what he was talking about!) Parallel methods do not in of themselves constitute what I call structural reasoning. I object to the funneling and flat methods of reasoning itself. Although I do not have any new alternatives to add to logic, fuzzy logic, probability, genetic algorithms and various network decision processes, my objection is directed toward the narrow focus on the fundamentals of those decision making processes, or to the creative (but somewhat dubious) steps taken to force the data to conform to the inadequacies of (what I called) flat decision processes. For instance, when it is discovered that probabilistic reasoning isn't quite good enough for advanced nlp, many hopefuls will rediscover the creative 'solution' of using orthogonal multidimensional 'measures' of semantic distance. Instead of following their intuition and coming up with ways to make the reasoning seem more natural, they first turn toward a more fanciful method by which they try to force the corpus of natural language to conform to their previously decision to use a simple metric. My recommendation would be to first try to begin thinking about how natural reasoning might be better structured to solve those problems before you start distorting the data. For an example, reasons are often used in natural reasoning. A reason can be good or bad. A reason can provide causal information about the reasoning but even a good reason may only shed light on information incidental to the reasoning. The value of a reason can be relative to both the reasoning and the nature of the supplied reason itself. My point here is that the relation of reason to reasoning is significant (especially when they work) although it can be very complicated. But even though the use of a reason is not simple, notice how natural and familiar it seems. Example: 'I do this because I want to!' Not a good reason to explain why I am doing something unless you are (for instance) curious about the emotional issues behind my actions. Another example: "I advocate this theory because it seems natural!" A much better reason for the advocacy. It tells you something about what is motivating me to make the advocacy but it also tells you something about the theory as it is being advocated. There are other kinds of structures to reasoning that can be considered as well. This was only one. I realized during the past few days, that most reasoning in a contemporary AGI program would be ongoing and so yes the reasoning would be more structured than I originally thought. (I wouldn't have written my original message at all except that I was a little more off than usual that night for some reason.) However, even though ongoing reasoning does represent some additional complexity to the process of reasoning, the fact that structural reasoning itself is not being discussed means that it is being downplayed and even ignored. So you have the curious situation where the less natural metric of semantic distance being enthusiastically offered while a more complete examination of the potential of using natural reasons in reasoning is almost totally ignored. So while I believe that modifications and extensions of logic, categorical systems, probability, and network decision processes will be used to eventually create more powerful AGI programs, I don't think the contemporary efforts to produce such advanced AGI will be successful without the conscious consideration and use of structural reasoning. Jim Bromer ------------------------------------------- agi Archives: https://www.listbox.com/member/archive/303/=now RSS Feed: https://www.listbox.com/member/archive/rss/303/ Modify Your Subscription: https://www.listbox.com/member/?member_id=8660244&id_secret=126863270-d7b0b0 Powered by Listbox: http://www.listbox.com
